PREC
PREC = Normal  Single  SingleN  Accurate  Low  Medium  High
Default: PREC  = Medium  for VASP.4.X 
= Normal  since VASP.5.X 
Description: PREC specifies the "precision" mode.
PREC sets default values for the energy cutoff ENCUT, the FFT grids (NGX,NGY,NGZ) and (NGXF,NGYF,NGZF), and the accuracy of the projectors in real space ROPT (used only when LREAL=.TRUE.). Details are given below in the table.
We recommend using PREC=Normal or PREC=Accurate. PREC=Normal can be used for most routine calculations. PREC=Accurate leads to a denser grid (NGX,NGY,NGZ). Thus, it reduces eggbox effects and strictly avoids any aliasing/wraparound errors. PREC=Normal and PREC=Accurate use an augmentation fine grid (NGXF,NGYF,NGZF) that is twice larger than the grid (NGX,NGY,NGZ) used for the representation of the pseudoorbitals. PREC=Accurate increases the memory requirements somewhat, but it should be used (in combination with an increased value for ENCUT) when a very good accuracy is required, e.g., for accurate forces, for phonons and stress tensor or in general when second derivatives are computed. The accuracy of forces can also be sometimes further improved by specifying ADDGRID=.TRUE., however, reports from users are somewhat contradictory about whether this really helps.
Mind:

Deprecated: The old settings PREC=Medium and PREC=High are no longer recommended and are available (as well as PREC=Low) only for backward compatibility. Essentially, PREC=High only increases the energy cutoff by 30 %, which can also be achieved by just manually increasing ENCUT. 
Important:

Default values set by PREC
Default values set by PREC for the parameters ENCUT, (NGX,NGY,NGZ), (NGXF,NGYF,NGZF) and ROPT:
PREC ENCUT NGX,Y,Z NGXF,YF,ZF ROPT (LREAL=A) ROPT (LREAL=O) Normal max(ENMAX) 3/2× 2×NGX 5×10^{4} 1.0 Single (VASP.5) max(ENMAX) 3/2× NGX 5×10^{4} 1.0 Single (VASP.6) max(ENMAX) 2× NGX 5×10^{4} 1.0 SingleN (VASP.6) max(ENMAX) 3/2× NGX 5×10^{4} 1.0 Accurate max(ENMAX) 2× 2×NGX 2.5×10^{4} 1.0 Deprecated settings: Low max(ENMIN) 3/2× 3× 1×10^{2} 2/3 Medium max(ENMAX) 3/2× 4× 2×10^{3} 1.0 High 1.3×max(ENMAX) 2× 16/3× 4×10^{4} 1.5
where max(ENMAX) and max(ENMIN) are the maxima of ENMAX and ENMIN found in the POTCAR file, and and are defined by
Further remarks
 With PREC=Normal, Single, and Accurate the grid (NGXF,NGYF,NGZF) representing the augmentation charges, charge densities and potentials has either the same size (PREC=Single) or the double size (PREC=Normal or Accurate) as the grid (NGX,NGY,NGZ). With the deprecated (and no longer recommended) settings for PREC (Low, Medium and High), the grid (NGXF,NGYF,NGZF) is determined by some heuristic formula from ENAUG.
 PREC=Single uses the same grid (NGX,NGY,NGZ) as PREC=Normal in VASP.5, but the same grid as PREC=Accurate in VASP.6. However, the double grid technique is not used, i.e., (NGXF,NGYF,NGZF)=(NGX,NGY,NGZ). This is convenient if one needs to cut down on storage demands or if one wants to reduce the size of the files CHG and CHGCAR. Furthermore, PREC=Single avoids highfrequency oscillations caused by the doublegrid technique and the resulting Fourier interpolation. It is often expedient for scanning tunneling simulations or other calculations where highfrequency wiggles of the charge density in the vacuum region are undesirable.
 PREC=Medium and PREC=Normal should usually be sufficiently accurate and lead to errors of less than 1 meV/atom.
 PREC=High should guarantee that the absolute energies are converged to a few meV and the stress tensor converged within a few kBar.
Related tags and articles
NGX, NGY, NGZ, NGXF, NGYF, NGZF, ENCUT, ENAUG, ENMIN, ENMAX, ROPT, LREAL, ADDGRID, PRECFOCK